LINGUIST List 24.98

Wed Jan 09 2013

Review: Computational Linguistics; Historical Linguistics; Linguistic Theories; Text/Corpus Linguistics: Steels (2012)

Editor for this issue: Rajiv Rao <>

Date: 09-Jan-2013
From: Nick Moore <>
Subject: Experiments in Cultural Language Evolution
E-mail this message to a friend

Discuss this message

Book announced at

Editor: Luc SteelsTitle: Experiments in Cultural Language EvolutionSeries Title: Advances in Interaction Studies 3Publisher: John BenjaminsYear: 2012

Reviewer: Nick Moore, Sheffield Hallam University


The ten papers collected in “Experiments in Cultural Language Evolution”represent the state-of-the-art of research into simulated multi-agentinteraction. Centered around Luc Steels’ work at the Sony Computer ScienceLaboratory, Paris, this volume represents the culmination of more than adecade of work dedicated to uncovering the practicalities of languageevolution in a social setting. The book is divided into three sections. Anintroductory section comprises a Foreword and an Introduction, both by Steels,that set out the direction and the theoretical framework for the remainingpapers. Part 1 describes experiments in vocabulary evolution and Part 2details how grammatical features evolve in experiments in the same framework.Each experiment enhances results gained in previous experiments.

The Foreword places the volume in its historical context by stressing that thequestion of language evolution is almost as plagued by speculation today as in1995, when Steels launched this research project. Because there is no fossilrecord and because we cannot allow any modern language to represent languagesas they first emerged, we can only be guided by general principles ofevolution when theorising the evolution of languages. Steels and his team havesince synthesised an approach to language evolution that attempts to simulatethe evolution of language in a cultural context by using computational agents,typically embodied as robots. The Foreword also summarises each chapter.

Chapter 1, “Self-organization and selection in cultural language evolution” bySteels, outlines the theoretical framework for the empirical descriptions inthe remaining chapters. Steels demands that any theoretical description oflanguage evolution be biologically feasible, demonstrate advantage to socialreproduction, and adapt to cultural change. Language in this model is assumedto be open-ended, distributed, and transmitted non-telepathically. The keyaspects of an evolutionary theory that are applied to language arefundamentally functional, i.e., Does language succeed in communicating? Agentsapply general strategies that adapt language for optimum expressive adequacy,cognitive effort, learnability and social conformity. The repeated applicationof these strategies to instances of communicative events produces a languagesystem based on the probability of communicative success. The language systemis the combination of the general cognitive capabilities of routine processingand meta-analysis. Ready-made responses may be available to a speaker, butanalysis is required to evaluate those responses. Where self-evaluationindicates a lack of success, a repair is introduced. Repair actions mayrequire a reframing of the chosen sentence, the selection of an alternativelexical item, or the creation of a new item or structure. Self-evaluation ispossible because of a routine termed ‘re-entry’ (i.e. a process that matchesthe mirror-neuron hypothesis; see Rizzolatti and Craighero, 2004), whichallows the speaker to practice the communicative effect of the chosen sentencebefore it is articulated by acting as the hearer in an internal process.

While the language system adapts, constrained by language strategies, languageitems emerge through a self-enforcing cumulative process of invention, trial,and alignment between agents. As with repair, alignment is central to theself-organising character of language. Alignment is the social enaction offrequency, such that the communicatively successful use of a language itemincreases the likelihood of it being adopted by other agents. This process isdemonstrated throughout the volume in various experiments. To furtherstrengthen the centrality of self-organisation in language evolution, Steelsalso employs the principle of 'structural coupling' (Maturana, 2002), whichfacilitates alignment through linguistic transmission due to the structure ofan organism and its interaction with the living, non-living and linguisticenvironment, without the need for intention or a central authority.

The key issue for Steels is to provide empirical evidence for the theoreticalframework sketched here. Contemporary evolutionary linguistic processes, suchas creolisation, can shed light on how language evolves, as can placinglinguistically-competent subjects into a context where new language must beinvented to complete a communicative task. However, Steels and hiscollaborators choose to model evolutionary processes computationally androbotically, using embodied agents to enact language games. Throughout thevolume, robots engage in: acquisition experiments, where one linguisticallycompetent robot passes on a linguistic system to another robot with a pidginversion of the language, through tutoring, although neither robot knows whichhas the full version; emergent experiments, where both robot agents, using thestrategies described above, collaborate to converge on a non-predeterminedstable linguistic system; and reconstruction experiments, where strategies arevaried by agents to simulate known linguistic evolution. The remainingchapters describe these experiments for selected vocabulary (Part 1) andgrammatical (Part 2) features.

Steels and Martin Loetzsch start Part 1 with the simplest language game: thenaming game. In the “non-grounded” version of this emergent experiment, twoagents share the same viewpoint of a set of objects. The speaker offers a namefor an object, to which the hearer points. If the hearer matches the speaker’sobject, a new round is played. However, a number of repairs may occur. Thespeaker may identify an object with no known name, in which case it has toinvent one. The hearer may not know the word, so it guesses the object. If theguess is correct, the new word is remembered, but if it is incorrect, thespeaker points out the object, and the new word is remembered. If the hearerknows a different word for the same object, scores are given to the differentwords so that, through usage, agents converge on agreed words. Thus, in oneexperiment, twelve words for five objects after 50 games become five to sixwords, on average, after 200 games. In the “grounded” version of the game, theagents are mobile and may see the same objects from different angles.Identifying objects through luminance, yellow/blue and red/green scores, x andy coordinates, and height and width measurements, agents store prototypes ofobjects which they then collaborate to name with other agents, using similarstrategies and repairs as in the previous game. Aggregate results produceclose to 100% communicative success after 1,000 games producing 20 terms after18 views of 10 objects. Adding the ability to both track moved objects andupdate prototype models results in about 90% success with 11 terms from 1,500games after 16 views of 10 objects. That is, these two learning heuristicsproduce far less ambiguity and synonymy.

In “Language Strategies for Color”, Joris Bleys engages robot agents in naminggames for colour, thereby accounting for how categories emerge from a naturalspectrum. Agents carry out the same language games as in the previousexperiment, but here the objects are distinguishable only by colour. Robotagents use a learning strategy that adjusts, rather than replaces, the currentprototypical colour towards the speaker’s use of the colour word whenevercommunication is unsuccessful. Using English words based on scores forbrightness, red/green and yellow/blue scores, robot agents score about 83%communicative success, matching baseline or target scores set by human agents.In an emergent experiment using only hue (or brightness), robot agents achieveabout 72% success. To make the experiments more closely match naturallanguage, Bleys also investigates graded membership of colour categories (e.g.“only slightly”, “somewhat” or “very” red). In a reconstruction experiment,robot agents produce words that were “qualitatively similar” (p.74) to theirhuman counterparts in baseline data. Similarly, in acquisition experiments,robot agents demonstrate communicative success at rates marginally belowhumans. An emergent experiment for colour produces almost 95% communicativesuccess with little variance for 5 words after about 15,00 games. Theimpressive results for graded membership demonstrate another important aspectof these evolutionary experiments: language strategies adapt to give selectiveadvantage. In this case, graded membership of colours allows a higher rate ofsuccess than brightness-only or hue-and-brightness systems.

The experiments in the next two chapters, “Emergent mirror systems for bodylanguage” by Steels and Michael Spranger and “The co-evolution of basicspatial terms and categories” by Spranger, add complexity to the linguisticmodels developed in the previous two chapters by adding verbal and adverbialoptions (Steel and Spranger) and prepositional meanings (Spranger). Spranger’sexperimental embodied-robotic subjects achieve 98% communicative success whenreconstructing German spatial terms. Steel and Spranger claim that “It is onlyby the full integration of all aspects of language with sophisticatedsensory-motor intelligence that agents were able to arrive at a sharedcommunicative system that is adequate for the game” (107) of correctlyordering a fellow robot agent to strike a particular pose. That is,communicative success is achieved by: grounding the agents in a sensoryexperience relative to their own body and its parts; employing a prototypical,rather than categorical, approach to language; simulating mirror neurons (byenabling robots to simulate and monitor, without enacting, a motor programme);and providing feedback loops for the motor system.

Part 1 culminates in the chapter “Multi-dimensional meanings in lexicalformation”, by Pieter Wellens and Loetzsch, which attempts to simulate a morenatural environment for lexical emergence and demonstrate the adaptivebenefits of the strategies adopted in the studies in this volume. The languagegames played by robot agents in the preceding chapters all focus on one aspectof language, but this does not reflect natural language use, when speakersmust select the most suitable linguistic features to distinguish objects. Themost favourable results are obtained when agents use a probability-based‘Adaptive Strategy’ for word learning, whereby a fuzzy-logic algorithm for‘best fit’ is used in naming objects as speaker or hearer. In experimentswhere 25 agents able to distinguish 16 features per object play 4,000 gameseach, totalling 50,000 games over 10 repetitions, the agents achieve 90%communicative success after 10,000 games, and approach a 98% success rateafter 30,000 games. Another measure, lexicon coherence, which quantifies thealignment between agents’ lexicons at any time, reaches 0.4 after 10,000 gamesand averages only as high as 0.45 on a scale of -1 to +1 after 50,000 games.This reflects natural language, where high levels of communicative success areachieved even when agents do not totally agree on word meanings.

Part Two of the book, ‘Emergence of Grammatical Systems’, opens with Remi vanTrijp’s ‘The evolution of case systems for marking event structure’, whichposits three bold hypotheses: 1. “Case evolves because it has selectiveadvantage for communication” (170); 2. case emerges when a population shares a‘case strategy’; and 3. “Case markers can be repurposed for a differentlanguage system if the original selective advantages of a case system havebeen ‘usurped’ by more dominant, competing systems in the language” (170). Inexperiments where one robot agent describes a scene that the two agents havejust watched together, robot agents acquire the case system for German,although van Trijp rejects the need for 'a priori' grammatical categories.After 5,000 games, coherence scores are above 0.95, the language system ishighly systematic, and cognitive effort is at a minimum, thus providingsupport for the first hypothesis. Moreover, the evolution of the Spanishpersonal pronoun system is reconstructed in experiments that provide evidencefor hypotheses 2 and 3 above. As with native speakers, grammatical variationis accommodated by robot agents who produce language with preferences forcertain structures. Similarly, subsequent experiments demonstrate a paradigmshift in the population, with preferences moving from one system to another.In the conclusion, van Trijp is careful to emphasise that these experimentsdemonstrate a high level of communicative success using general sharedcognitive strategies – typically, “analogical reasoning or similarity-basedcategorization” (202).

In “Emergent functional grammar for space”, Spranger and Steels demonstratethe selective advantage of grammaticalising spatial relationships over thesolely lexical variant in experiments that reconstruct German and thatself-organise into an emergent system. Crucially, they show how asemantically-oriented strategy towards grammaticalised spatial relationshipsrequires less cognitive effort for greater communicative success. Similarly,Katrien Beuls, Steels and Sebastian Höfer’s experiments into “The emergence ofinternal agreement systems” produce results that reduce cognitive effort andambiguity by grouping related words into groups or phrases. KaterynaGerasymova, Spranger and Beuls investigate the Russian system of Aktionsartenin “A language strategy for aspect”. Although the Russian system of aspect isconsidered complex and elaborate, robot agents are able to reconstruct andacquire the system, partially aided by the ability to accept holophrases (alearned combination of words) for later analysis. Robot agents thendemonstrate how an entirely new aspect system can emerge. As in theexperiments by Wellens and Loetzsch, the final chapter ''The emergence ofquantifiers'', by Simon Pauw and Joseph Hilferty, demonstrates the selectiveadvantage of fuzzy categories by focusing on quantifying expressions.Experiments in acquisition and formation compare the alternative strategies ofabsolute quantification and scalable quantification, resulting in theconclusion that the more unpredictable the environment, the more likely ascalable strategy will prevail.


Although each paper has different authors, the volume exhibits both aremarkable sense of consistency and a clear sense of progression from onechapter to the next. The research reveals a sense of direction shared bySteels and the other contributors that is laid out in Chapter One. In fact, itis advisable to read Chapter One again after examining the results of laterexperiments, in order to fully appreciate the significance of the boldapproach taken by this team of researchers.

The greatest danger of depending on functional explanations to support ahypothesis is that evidence can only be interpreted as supporting an inertstatus quo. Fortunately, Steels and colleagues avoid this theoretical blindalley by incorporating the dynamics of alignment and the explanations andmechanisms for linguistic change. For instance, in van Trijp’s chapter,experimental evidence provides support for the hypothesis that the advantagesprovided by grammatical case in Spanish have been replaced by othergrammatical features, freeing case markers to function in new ways. Perhaps myonly concern with some of the papers in the volume is that there is anover-reliance on formal, rather than functional models of language. While somefunctional models may be difficult to model computationally, there aresolutions, such as Halliday and Matthiessen (1999), which may provide theresearch team with grammatical models more aligned with thenon-representational approach to language that is central to the researchreported here.

This book and other experiments by the same team provide empirical evidencefor the emergence of language based on evolutionary principles, on what wecurrently understand about brain structure and organisation (e.g. Edelman1999; 2004) and, significantly, without the need for language-specificacquisition strategies; in all of the experiments here, the learningstrategies employed are general cognitive strategies rather thanlanguage-specific. The experiments repeatedly demonstrate that: language canemerge without a priori conditions; current language systems can be alignedwithin a community through structural coupling; known developments in languagecan be modelled in embodied robotic agents with simulated mirror neurons; andlanguage functions probabilistically, not categorically. I am unaware of anyother series of falsifiable experiments that provide verifiable evidence tocounter these conclusions, despite many theoretical claims to the contrary.Consequently, this volume should be of value to anyone interested in languageevolution, in the application of natural languages to robotic agents, and ingeneral linguistic theory.


Edelman, G.M. 1999. Building a picture of the brain. Annals of the New YorkAcademy of Sciences 882 June 1999, p.68-89

Edelman, G.M. 2004. Wider Than the Sky - The Phenomenal Gift of Consciousness.New Haven: Yale University Press

Halliday, M.A.K. and Matthiessen, C.M.I.M. 1999. Construing Experience throughMeaning: A Language-based Approach to Cognition. London: Continuum

Maturana Romesin, H. 2002. Autopoiesis, Structural coupling and cognition: Ahistory of these and other notions in the biology of cognition. Cyberneticsand Human Knowing 9(4), pp.5-34

Rizzolatti, G. and Craighero, L. 2004. The Mirror-Neuron system. Annual Reviewof Neuroscience 27, pp.169-92

About the Reviewer:Nick Moore has worked in Brazil, Oman, Turkey, the UAE and the UK withstudents and teachers of English as a foreign language, English for specificand academic purposes, and linguistics. His PhD in applied linguistics fromthe University of Liverpool addressed information structure in writtenEnglish. His other research interests include systemic functional linguistics,corpus linguistics, theories of embodiment, lexis and skills in languageteaching, and reading programs. He is the co-editor of 'READ', maintains ablog on language, linguistics and learning at and hasrecently joined the TESOL unit at Sheffield Hallam University.

Page Updated: 09-Jan-2013